Artificial Intelligence and Academic Detailing

Jerry Avorn, M.D.
Co-Founder & Special Adviser, NaRCAD

​Tags: Evidence Based Medicine, Jerry Avorn​

​Following the astonishing debut of AI applications like ChatGPT a year ago, “knowledge workers” (that’s us) have been forced to ponder how much of what we do could be replaced by a very smart set of computer programs. Such applications can already pass medical licensing exams better than many graduates and have gotten remarkably good at reading X-rays and pathology specimens. How soon will AI systems become adept at reviewing the clinical literature and preparing concise, user-friendly summaries, complete with prescribing recommendations?  Not yet, but likely before long.

---

​Try it yourself at home: log onto OpenAI.com (it’s free) and ask ChatGPT for advice about medications for diabetes or hypertension or HIV or anything else.  Just be careful about its “hallucinations” – the fact that sometimes AI just makes up wrong stuff. (I prefer the term “confabulation,” also used to describe this well-known phenomenon.) That can be whimsical if you’re a N.Y. Times reporter and ChatGPT advises you to leave your spouse, and it can be very problematic if you’re a lawyer who relies on case law that ChatGPT simply fabricated. (Both actually happened.) But it can be lethal if it involves incorrect clinical recommendations.

​Yet that said, AI is getting smarter every day. If programmed well in the coming years, large language models like ChatGPT or its growing number of competitors could eventually also learn how to gauge prescribers’ current knowledge, attitudes, and practices, and then ask just the right questions to find out why they’re doing what they’re doing, what their concerns are, and what it would take to get them to change.

---

Once things mature a bit further, will large health care systems interested in academic detailing and in cost-cutting simply replace humans with AI-AD-bots? After all, they could work 18-hour days, don’t need health care benefits, and can disseminate any message their employer wants. It will be easy replace a recommendation like “SGLT-2 inhibitors in diabetes can reduce cardiovascular and renal disease as well as lower glucose” with: “SGLT-2 inhibitors are extremely expensive and increase our drug budget.  Use metformin or sulfonylureas whenever possible.

---

​So if we have a few years to prove that actual people still have a vital role to play in helping practitioners make better decisions, what can we do? 
​

• First, we should consider and then embrace what makes us fully human in our work: our capacity to bring a critical eye to the clinical literature, and to understand and embrace nuances in the evidence.
• Second, we can draw on our genuine empathy for the perspective of the practitioner, and an appreciation that that it’s hard to be a front-line clinician. This is a property the bots may eventually be able to simulate, but at the moment still do imperfectly.
• Third and most important, we can remember to bring to our work a clear moral mission – that good evidence matters a lot, and that patients have to come first.

​Those are values that endure and can distinguish our work from a sophisticated set of algorithms. Best of all, they can’t be changed if whoever is in charge overwrites a few lines of code to maximize some other agenda, or if the algorithms just make stuff up.

---

Biography. 
Jerry Avorn, MD, Co-Founder & Special Adviser, NaRCAD
Dr. Avorn is Professor of Medicine at Harvard Medical School and Chief Emeritus of the Division of Pharmacoepidemiology and Pharmacoeconomics (DoPE) at Brigham & Women's Hospital. A general internist, geriatrician, and drug epidemiologist,  he pioneered the concept of academic detailing and is recognized internationally as a leading expert on this topic and on optimal medication use, particularly in the elderly. Read More.